2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
SSDs in the Cloud
Dave Wright, CEOSolidFire, Inc.
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Storage in the Cloud
2
Loca
l Sto
rage • Swap
• Temp Files• Data
Processing
• WD• Seagate• FusionIO
Bloc
k St
orag
e • VM Images• Application
Files• Databases
• EMC• Netapp• 3par• SolidFire
Bulk
Obj
ect
Stor
age • Media Files• Content
Distribution• Backup• Archival
• S3• Atmos• Nirvanix
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Block Storage Challenges
Performance IOPS/GB balanceMulti-tenant isolation (seek latency)
EfficiencyUtilizationPower/cooling
ManagementAutomationReliability
3
Storage IO
Server IO
time
IO
Usable
Purchased
Storage Capacity
Usable
Administration Interface
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Flash Benefits for Block Storage
PerformanceRestore balance between IOPS and CapacityBetter performance with varied workloads
EfficiencyRemove performance as a limiter on utilizationReduce storage space, power, and cooling
ManagementBetter reliability?
4
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
What’s the downside?
CostPartially a perception issueCan’t be ignored – cloud all about value prop
Storage ArchitecturesAbility to get full performance from flashWrite amplification & endurance issues
5
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
How to use Flash today
Direct-attached Host based caching Array based caching Array based tiering All-solid-state arrays
6
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Direct Attached
Examples:FusionIO, Virident, 2.5” SSD
Pros:Best performance
Cons:Limited capacity / flexibilityNo sharingLimited availability / redundancy
7
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Host based caching
Examples:FB FlashCache, IO Turbine, Adaptec, Marvell
Pros:Larger flash footprintLower latency than array based caching
Cons:Read-only, or affects data integrity /
availabilityHuge delta in performance between
cached/uncached reads8
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Array based caching
Examples:Netapp FlashCache, EMC FAST Cache
Pros: Invisible / seamless improvementProvides good boost for hottest data
Cons:Requires expensive SLC flash due to churnCan cause irregular performance in multi-tenant
environment
9
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Array based tiering
Examples:EMC FAST, 3par, Compellent
Pros:Potentially large capacity availableAbility to automate or manually place data in tiers
Cons:Requires expensive SLC flash due to churnController overhead moving data back and forth Irregular performance when automated
10
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
All-Solid-State arrays
Examples: SolidFire, Nimbus, Violin, TMS, Pure
Pros:High performance for all dataBest $/IOPPotential for less power/cooling expense
Cons:Cost/GB?Designed for cloud scale?
11
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
SolidFire Solution Architecture
Clustered Scale-Out Storage Solution Leverages Industry-Standard Hardware Native 10 GigE iSCSI All Solid State (SSD)
Combine 100 nodesOver 5M IOPS
Usable $/GB similar to traditional SANs
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Bridging the Performance Gap
SolidFire – Guarantee Storage Performance 50,000 IOPS/node. 5,000,000 IOPS/cluster. System-wide Performance Virtualization
Guarantees sustained performance (QoS) to every volume Enables volume-level performance throttling Create and bill against performance tiers Real-time performance adjustments via API
SolidFire Helix™ Data Protection – manages device failures without performance impact
Storage IO
Server IO
time
IO
Increasing gap between Server and Storage IO• Increasing latency, unhappy customers• Unable to guarantee performance• Performance degrades with disk / device failure
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
New Level of Storage Efficiency
Usable
Purchased
Storage Capacity
Inefficient use of purchased capacity• Deduplication and Compression unavailable or turned off
to maintain performance• Capacity Sprawl = large power and cooling cost
SolidFire – Operational Savings High capacity utilization - 85% Proactive inline Deduplication, Compression and Thin Provisioning
Executed across the entire data store, w/o performance impact Deduplication gains (30% - 70%) Compression gains (30% - 70%)
Drastic reduction in power, cooling, and floor space
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Cloud Specific Management
SolidFire Advantage REST-based API (management and automation)
Designed around the cloud service providers’ needs to automate storage management and provisioning
Extremely simple to build user-facing storage controls with your own business rules
Enables multi-tenant provisioning, reporting, and performance isolation Any SolidFire function can be automated via API
Usable
Administration Interface
Proprietary Management• Complex, Individual GUI interface• Manual deployment and provisioning• Lack of “cloud specific” API capability
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Volume
Metadata
Block D
ata
LUN 1
LUN 2
VolumesLUN 3
SolidFire
SolidFire Architecture
16
LUN 1
LUN 2
LUN 3
Disk Pool
Volumes
LUN 1
LUN 2
LUN 3
SAN 1.0
LUN 1
LUN 1
LUN 1
LUN 2
LUN 3
LUN 3
LUN 2
LUN 2
Disk Pool
Volumes
SSD Pool
LUN 3
SAN 2.0 Completely separate data storage from higher level representations
Allocate physical space in byte-size chunks
Optimize storage of small blocks Utilize entire drive pool for
every volume Enables:
Performance virtualization De-dupe and compression without
tradeoffs HA without performance impact
2011 Storage Developer Conference. © SolidFire, Inc. All Rights Reserved.
Summary
Flash can help overcome key primary storage issues in the cloud
Many possible approaches to using it If the right techniques are used to reduce $/GB, all-
solid-state solutions are viable Future direction of cloud storage is clear: Disk for capacity oriented, secondary storage Flash / SSD for all primary storage
17